Ethical Large-Scale Artificial Intelligence




Ethical Large-Scale Artificial Intelligence

3 April 2019


Added 02-May-2019

Unintended bias and unethical Artificial Intelligence (AI) technologies can be detected by fairness metrics and corrected with mitigation techniques. Fair computational intelligence is important because AI is augmenting human tasks and decisions within every facet of life. As a core component of society, sports and entertainment are becoming driven with machine learning algorithms. For example, over 10 million ESPN fantasy football players use Watson insights to pick their roster week over week. A fair post processor ensures NFL players, irrespective of the team assignment, are projected for an impartial boom in play so that owners avoid basing their team roster decisions on biased insights. This is critically important because users spent over 7.7 billion minutes on the ESPN Fantasy Football platform during the 2018 season. In another example, automated video highlight generation at golf tournaments should be contextually fair. Golf player biographical data, game play context and weather information should not skew deep learning excitement measurements. An overall player video highlight excitement score that includes gesture, crowd noise, commentator tone, spoken words, facial expressions, body movement and 40 situational features is continually debiased. The resulting highlights are pulled into personalized highlight reels and stored on a web accelerator tier. Throughout the talk, I will show examples of using an open source library called IBM AI Fairness 360 and the IBM OpenScale cloud service to provide highly veracious insights.